TSRE: Channel-Aware Typical Set Refinement for Out-of-Distribution Detection

TL;DR

This work tackles the challenge of reliable OOD detection by addressing limitations of activation-based rectification, notably the neglect of channel-specific statistics and distributional skew. It introduces TSRE, a channel-aware typical-set refinement that jointly leverages per-channel discriminability, activity, and skewness to adaptively shape typical activations before computing an energy-based OOD score. Through prototypes-based channel metrics, adaptive per-channel boundaries, and skewness-corrected refinement, TSRE achieves state-of-the-art performance on ImageNet-1K and CIFAR-100 across multiple backbones and OOD scoring functions. The approach demonstrates strong generalization and robustness, with ablation studies confirming the essential roles of discriminability, activity, and skewness in boosting ID-OOD separability.

Abstract

Out-of-Distribution (OOD) detection is a critical capability for ensuring the safe deployment of machine learning models in open-world environments, where unexpected or anomalous inputs can compromise model reliability and performance. Activation-based methods play a fundamental role in OOD detection by mitigating anomalous activations and enhancing the separation between in-distribution (ID) and OOD data. However, existing methods apply activation rectification while often overlooking channel's intrinsic characteristics and distributional skewness, which results in inaccurate typical set estimation. This discrepancy can lead to the improper inclusion of anomalous activations across channels. To address this limitation, we propose a typical set refinement method based on discriminability and activity, which rectifies activations into a channel-aware typical set. Furthermore, we introduce a skewness-based refinement to mitigate distributional bias in typical set estimation. Finally, we leverage the rectified activations to compute the energy score for OOD detection. Experiments on the ImageNet-1K and CIFAR-100 benchmarks demonstrate that our method achieves state-of-the-art performance and generalizes effectively across backbones and score functions.

Figures (4)

• Figure 1: Significant deviation between the actual activation and the Gaussian assumption.
• Figure 2: Ablation study on TSRE. The distribution overlap between ImageNet-1K (ID) and SUN (OOD) increases when each channel-aware factor is removed, validating the contributions of discriminability, activity, and skewness to enhancing ID-OOD separability.
• Figure 3: Sensitivity analysis of hyperparameters $\omega$, $\theta$, and $p$ on FPR95 and AUROC.
• Figure 4: The distribution of scaling factors $\lambda_i$ after adaptive refinement.